You are deciding which vector database to put under your internal AI assistant, and the finance team wants a number before the project goes any further.
Here is the difficulty. All three of the leading managed vendors changed their pricing in the last twelve months, none of them price the same unit, and only one of them publishes a figure you can put straight into a budget line.
This page sets out what Pinecone, Qdrant and Weaviate actually charge as of 13 August 2026, taken from each vendor's own published pricing, along with the costs that do not appear on any pricing page.
What does a vector database actually cost in 2026?
Entry paid plans start at US$20 to US$50 per month across the three leading managed vendors. Enterprise-grade tiers with an uptime SLA, single sign-on and private networking start at US$280 to US$500 per month as a minimum commitment, and rise with usage. None of the three caps your bill at the published minimum.
The critical distinction for budgeting is that these are floors, not forecasts. Pinecone Standard and Enterprise are minimum monthly spends applied against usage. Weaviate Flex and Plus are starting prices. Qdrant publishes no dollar figure at all above the free tier.
Published entry prices at a glance
--- Pinecone Starter: free, AWS us-east-1 only
--- Pinecone Builder: US$20 per month, flat
--- Pinecone Standard: US$50 per month minimum usage
--- Pinecone Enterprise: US$500 per month minimum usage
--- Weaviate Flex: from US$45 per month, pay as you go
--- Weaviate Plus: from US$280 per month
--- Weaviate Premium: minimums vary by configuration, not published
--- Qdrant Free Tier: free forever, 0.5 vCPU, 1 GB RAM, 4 GB disk
--- Qdrant Standard: usage-based hourly billing, no published rate card
--- Qdrant Premium: minimum spend required, amount not published
How much does Pinecone cost?
Pinecone publishes four plans. Starter is free, Builder is US$20 per month flat, Standard carries a US$50 per month minimum usage commitment, and Enterprise carries a US$500 per month minimum. Above each minimum, billing is pay as you go against database, inference and assistant usage.
Pinecone is the only one of the three that publishes a hard number for its enterprise tier, which makes it the easiest to take to a budget meeting.
What each Pinecone plan includes
--- Starter (free): AWS us-east-1 only, up to 5 indexes, 2 GB storage, up to 2 million write units and 1 million read units per month, 1 GB egress, 1 project, up to 2 users, community support via Discord.
--- Builder (US$20 per month flat): everything in Starter with higher limits, choice of cloud and region, multiple projects and users, and Prometheus and Datadog monitoring.
--- Standard (US$50 per month minimum): adds Dedicated Read Nodes, import from object storage, backup and restore, role-based access control, SAML 2.0 single sign-on, and a HIPAA add-on. The trial runs three weeks and includes US$300 of credits.
--- Enterprise (US$500 per month minimum): adds a 99.95% uptime SLA, bring your own cloud, private endpoints, customer managed encryption keys, audit logs, service accounts, SAML roles, SCIM, HIPAA compliance and Pro support.
Pinecone also offers Committed Use Contracts, where larger usage commitments unlock discounts. Those rates are negotiated and not published.
How much does Weaviate Cloud cost, and what changed?
Weaviate Cloud runs three plans: Flex from US$45 per month, Plus from US$280 per month, and Premium with minimums that vary by configuration. This structure took effect on 27 October 2025, replacing the previous model in which Serverless Cloud became Shared Cloud and Enterprise Cloud became Dedicated Cloud.
The change is worth understanding because it raised the entry price for some existing users.
Under the old model, a non-highly-available cluster started at US$25 per month and a highly available cluster at US$75. Under the new model, every Shared cluster is highly available by default at an entry price of US$45. If you were running a non-HA cluster for US$25, your floor rose by 80%.
Weaviate now bills on three dimensions rather than one:
--- Vector dimensions: object count multiplied by vector dimensionality, multiplied again by your replication factor. Pricing varies by index type, HNSW or Flat, and by compression method.
--- Storage: total disk used for vector indexes, metadata, object properties and database state, in GiB per month.
--- Backups: total size of backed-up collections multiplied by your retention period.
List prices vary by cloud provider and region. Flex includes RBAC and 99.5% uptime; Plus adds annual commitment options, stronger SLAs, a choice of Shared or Dedicated deployment and 99.9% uptime; Premium is dedicated infrastructure at 99.95%.
The replication factor multiplier is the line most teams miss. Running three replicas for availability multiplies your vector dimension charge by three.
How much does Qdrant Cloud cost?
Qdrant publishes a permanently free tier and three paid tiers, but no dollar rate card. The free tier gives a single-node cluster with 0.5 vCPU, 1 GB RAM and 4 GB disk. Standard is usage-based hourly billing on vCPU, RAM, disk, backup storage and paid inference tokens. Premium requires a minimum spend that Qdrant does not publish.
What Qdrant does publish clearly is the service difference between tiers, which is often the real decision.
--- Standard: dedicated resources, vertical and horizontal scaling, backup and disaster recovery, 99.5% uptime SLA, support 10 hours a day on business days, severity 1 response within 4 business hours.
--- Premium: single sign-on, private VPC links, 99.9% uptime single availability zone or 99.95% multi-AZ, 24 by 7 support, severity 1 response within 1 hour, and an optional forward deployed engineer.
--- Hybrid Cloud: Qdrant's management plane operating clusters inside your own infrastructure, which is the relevant option where data residency is a licence condition rather than a preference.
--- Private Cloud: fully isolated deployment with custom SLAs, intended for air-gapped environments.
Qdrant carries SOC 2, GDPR and HIPAA attestations, and clusters run on AWS, Azure or Google Cloud from the free tier upward.
What is the annual floor for each option?
Multiplying published minimums by twelve gives the annual floor before any usage. Pinecone Standard is US$600 a year and Enterprise US$6,000. Weaviate Flex is US$540 a year and Plus US$3,360. Qdrant Standard has no floor at all, and Qdrant Premium has one that is not disclosed.
These figures are the smallest number the vendor will accept, not an estimate of your bill. Treat them as the entry ticket and model usage separately.
For a Hong Kong company of 200 to 500 staff putting an assistant over an internal document set, the practical planning range for the database line alone is roughly US$600 to US$8,000 a year, depending entirely on whether your compliance requirements force you into a tier with single sign-on, private networking and an uptime SLA.
That last clause is what actually sets the price. The vector database is rarely expensive because of data volume. It becomes expensive because of security and availability requirements.
Which costs are not on the pricing page?
Four costs sit outside the vector database bill and routinely double the real total: embedding generation, re-indexing when you change embedding models, egress charges when the database and application sit in different clouds, and the engineering time to maintain the ingestion pipeline.
Embeddings. Every document must be converted to vectors before storage, and every query must be converted before search. That is a separate line item on a separate vendor's invoice unless you use the database's bundled inference.
Re-indexing. Changing embedding model means regenerating every vector in your corpus. Budget for this at least once. Model deprecation makes it eventually unavoidable, a risk we set out in AI model deprecation.
Replication and backups. On Weaviate these are explicit billing dimensions. On the others they are absorbed into resource usage. Either way, three replicas cost roughly three times one.
Pipeline maintenance. Documents change. Somebody has to keep the index in step with the source systems, and that recurring engineering effort is usually larger than the database subscription.
Which vector database should you choose?
Choose Pinecone if you need a published enterprise price and the shortest path to procurement approval. Choose Weaviate if you want billing you can model in a spreadsheet before committing. Choose Qdrant if data residency or a hybrid deployment inside your own cloud is a hard requirement. Choose none of them if your corpus is small.
--- If you need a number for the board this quarter: Pinecone. US$500 per month minimum for the enterprise tier is the only firm enterprise figure published by any of the three.
--- If your finance team wants to model cost before signing: Weaviate. Objects, dimensions, replication factor, storage and backup retention are all inputs you control and can forecast.
--- If regulation requires the data to stay in your infrastructure: Qdrant Hybrid Cloud or Private Cloud. This is the clearest published path of the three for an air-gapped or in-network deployment.
--- If you need 24 by 7 severity 1 support: Qdrant Premium offers a 1 hour severity 1 response, but you will have to request the price.
--- If you have fewer than roughly 100,000 documents: consider not buying a vector database at all. A pgvector extension on the PostgreSQL instance you already run will handle that scale, and it does not add a vendor to your risk register.
What these vendors will not tell you
The published minimum is a floor, not a ceiling, and the honest answer to "what will this cost us" is that no vendor can tell you until your corpus size, query volume and replication requirements are fixed. Anyone quoting a firm annual figure before that has guessed.
Three further limitations belong in your evaluation notes.
Two of the three do not publish enterprise rates. Qdrant Premium and Weaviate Premium both require a sales conversation. If your procurement process needs list pricing to open a file, that is a real friction cost, not a detail.
Bundled inference creates a switching cost. Using a vendor's own embedding models is convenient and makes migration harder, because leaving means re-embedding your entire corpus with a different model.
None of this fixes retrieval quality. A more expensive database does not make your assistant more accurate. Retrieval quality is a function of chunking, embedding choice and evaluation, which is why we would rather you read what AI evals are before you compare price tables.
And UD's own limits. We are not a reseller of Pinecone, Qdrant or Weaviate, we hold no discount arrangement with any of them, and we do not earn a margin on your subscription. If you have one clear use case, a small document set and an engineer who has built this before, buy direct and skip the implementation partner. Where we add value is the harder part, which is deciding whether you need this layer at all, sizing it honestly, and connecting it to systems that were not designed with AI in mind.
The next step
Before you compare pricing tables again, fix three numbers: how many documents, how many queries per day, and whether single sign-on and private networking are contractual requirements or preferences. Those three answers select the tier for you, and until you have them, every quote you receive is a guess dressed as a proposal.
Then run the smallest real workload you can, on a free tier, with your own documents. A week of that will tell you more than any comparison table, including this one.
Getting this right is less about picking a vendor and more about knowing what your organisation actually needs before anyone quotes you. We understand AI. We understand you. With UD by your side, AI never feels cold.
Reviewed by the UD enterprise AI team, Hong Kong. Prices verified against vendor published pricing on 13 August 2026 and subject to change.
Not sure which layer you actually need?
Most organisations get quoted infrastructure before anyone has confirmed they need it. UD will walk you through every step, from sizing your document set and choosing the retrieval approach, to deployment, security review and cost modelling, with 28 years of Hong Kong enterprise experience behind it.
